<p align="center"> <img src="./assets/antislop-banner.png" alt="antislop" width="100%" /> </p> <p align="center"> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-2ea44f" alt="License: MIT"></a> <a href="https://github.com/miqdadbadjuber/anti-slop/releases"><img src="https://img.shields.io/github/v/release/miqdadbadjuber/anti-slop?label=version&color=1f6feb" alt="Version"></a> </p> <p align="center"> <a href="https://skills.sh/miqdadbadjuber/anti-slop"><img src="https://skills.sh/b/miqdadbadjuber/anti-slop" alt="skills.sh"></a> </p>

antislop

Anti Slop: Rules for AI Coding Agents. It stops them from generating generic "AI slop" UI and copy, without letting the result turn sterile. It is a filter, not a style guide: no prescribed colors, fonts, or layouts. It is not only for building pages: it also writes and audits copy, so AI text stops reading like AI. And it never beautifies on its own; DESIGN.md (yours) is where beauty and direction come from.

New here? Start with the GUIDE.md. It explains what antislop is and how to install it, from zero.

What it does

• 38 mandatory rules (R-01 to R-38) in three tiers: Hard Gate (absolute), Purpose-Gate (technique allowed, reason required), Quality Locks (consistency)
• A Liveliness Toolkit with three dials (ENERGY / RHYTHM / MOTION) and a Design Read, so the result is alive and specific, not just "clean"
• A Delivery Gate: a mandatory PASS/FAIL report in four blocks, run before anything ships
• Additive skills, one per concern, so an agent only loads what a task needs

The core prevents slop but cannot invent direction. DESIGN.md (yours) supplies it; a sterile result means the direction was missing, not that the filter failed (R-37).

See the difference

The same brief, the same page, generated four times over. antislop filters; DESIGN.md supplies direction. They are two different jobs, and these are the four results. Copy and code follow as before-and-after pairs instead of four separate builds, because nothing else has to supply direction there.

UI

Three of the four builds. The first uses neither tool, the second only antislop, the third only DESIGN.md.

| Nothing at all | antislop alone | DESIGN.md alone | |:--|:--|:--| | <a href="assets/compare/ui/without.webp"><img src="assets/compare/ui/without.webp" alt="A generic landing page: a sparkle logo, a NEXT-GEN AI 2.0 beta pill above the headline, and a fake terminal reporting 0.0001ms latency" width="100%"></a> | <a href="assets/compare/ui/w-antislop.webp"><img src="assets/compare/ui/w-antislop.webp" alt="The same page with antislop: honest copy on a restrained dark layout with a single accent colour" width="100%"></a> | <a href="assets/compare/ui/w-design.webp"><img src="assets/compare/ui/w-design.webp" alt="The same page with DESIGN.md only: a photographic hero, with the stat cards still reading 10,000% ROI Synergy Multiplier and a 5.0 rating from 500,000 founders" width="100%"></a> | | Where most AI output starts: a sparkle logo, a beta pill, and a fake terminal. | Honest, because the filter removed the invented numbers. Plain, because beauty is not its job. | The direction lands, but the slop stays, because DESIGN.md directs and does not filter. |

The fourth uses both, and it is the only one of the four that is clean and directed at the same time:

| antislop + DESIGN.md | |:--| | <a href="assets/compare/ui/w-all.webp"><img src="assets/compare/ui/w-all.webp" alt="The same page with antislop and DESIGN.md: a full-bleed illustrated hero with one honest headline and project-specific navigation" width="100%"></a> | | Honest numbers and a real direction in the same build. The filter removes what should not be there; DESIGN.md fills the space that leaves, which is the one thing neither tool manages alone. |

---

Copy

One prompt, run twice.

| Before | After | |:--|:--| | <a href="assets/compare/text/before.webp"><img src="assets/compare/text/before.webp" alt="An AI written Discord launch post: emoji bullet points, NEW DROP in capitals, and hype in every line" width="100%"></a> | <a href="assets/compare/text/after.webp"><img src="assets/compare/text/after.webp" alt="The same launch post written with antislop-copywriting: plain sentences, no emoji bullets, and a note to cut any line with nothing real to say" width="100%"></a> |

---

Code

One file, run twice.

| Before | After | |:--|:--| | <a href="assets/compare/code/before.webp"><img src="assets/compare/code/before.webp" alt="Python with box-drawing section banners, emoji, and a comment on every constant that restates the constant" width="100%"></a> | <a href="assets/compare/code/after.webp"><img src="assets/compare/code/after.webp" alt="The same Python with the banners and emoji gone, one comment left that says what the module does, and the code itself untouched" width="100%"></a> |

Every image in this section opens full size if you click it.

Install

antislop ships as a set of standard agent skills (one folder per skill, holding a SKILL.md). The core is always loaded; the other skills load only when the task needs them. This section is the reference: every install command, in one place. GUIDE.md walks through them from zero, and its Update and Remove sections cover every route the same way.

1. The installer (recommended)

One command, then answer the prompts. It asks which extra skills you want, where to install (this project or everywhere), and which agents you use, then copies the folders. It is also the only route that writes the pointer reloading antislop every session on a project install; a global install relies on the skills loading themselves by description.

npx antislop-ai

To update later, run the same command and choose Overwrite them when it finds the existing folders. See Update.

2. The skills directory

antislop is listed on skills.sh, the open directory for agent skills:

npx skills add miqdadbadjuber/anti-slop

This copies the same folders as path 1 and nothing else: no pointer, so antislop reloads by description alone. GUIDE.md covers adding the pointer afterwards, and updating or removing this route.

3. The plugin (Claude Code)

Add the marketplace once, then install the plugin:

/plugin marketplace add https://github.com/miqdadbadjuber/anti-slop
/plugin install antislop@anti-slop

4. The plugin (Antigravity)

The same repo is an Antigravity plugin. Install it with the Antigravity CLI:

agy plugin install https://github.com/miqdadbadjuber/anti-slop

5. The plugin (Codex)

The same repo is a Codex plugin and marketplace. Add the marketplace once, then install the plugin:

codex plugin marketplace add miqdadbadjuber/anti-slop
codex plugin add antislop@anti-slop

6. The plugin (Cursor)

The same repo is a Cursor plugin. Add it as a plugin marketplace with the Cursor Agent CLI:

agent plugin marketplace add https://github.com/miqdadbadjuber/anti-slop

Then open Customize in Cursor, find antislop, and select Install, choosing project or user scope.

7. The plugin (Kimi Code)

The same repo is a Kimi Code plugin. Install it in a Kimi Code session, then start a new one:

/plugins install https://github.com/miqdadbadjuber/anti-slop

The URL resolves to the latest release. Kimi Code installs plugins per user, so this covers every project.

8. The plugin (Cline)

The same repo is a Cline plugin. Install it with the Cline CLI:

cline plugin install https://github.com/miqdadbadjuber/anti-slop.git

The plugin ships no tools and no hooks. Its whole payload is the skills/ folder it bundles, which Cline discovers on install. This has to be said plainly: Cline's own documentation limits plugins to the SDK, the CLI, and Kanban, and says the feature does not apply to the VS Code and JetBrains extensions yet. So this door covers a Cline CLI install and not an editor install, and the installer in path 1 is the route for an editor.

9. The package (Pi)

The same repo is a Pi package. Pi reads a pi key in this repository's root package.json, which points at the skills/ folder, so the install is one command:

pi install git:github.com/miqdadbadjuber/anti-slop

That writes the package declaration to ~/.pi/agent/settings.json, which covers every project. Add -l to write it to this project's .pi/settings.json instead, where it applies to this repository only and loads after you grant project trust.

10. The plugin (Oh My Pi)

The same repo is an Oh My Pi plugin and marketplace. Oh My Pi reads the catalog this repository keeps at .omp-plugin/marketplace.json, so the only Oh My Pi file here is that catalog: the plugin's manifest is the .claude-plugin/plugin.json the Claude Code door already ships, and nothing is copied into your project. Add the marketplace once, then install the plugin:

omp plugin marketplace add miqdadbadjuber/anti-slop
omp plugin install antislop@anti-slop

Or from inside an active Oh My Pi session:

/marketplace add miqdadbadjuber/anti-slop
/marketplace install antislop@anti-slop

The install is per user by default, which covers every project. Add --scope project to keep it inside the active project instead.

Where the skills live

Every skill is a folder of the open Agent Skills standard (<name>/SKILL.md), so it drops into any agent that reads the standard. The installer (path 1) installs into whichever of these you use, creating the folder if it is missing:

| Agent | Reads antislop from | |-------|---------------------| | Claude Code | .claude/skills/ | | Codex | .codex/skills/ | | Antigravity | .agents/skills/ | | OpenCode | .opencode/skills/ | | Cursor | .cursor/skills/ | | Cline | .cline/skills/ | | Amp | .agents/skills/ | | Gemini CLI | .gemini/skills/ | | Hermes | .hermes/skills/ | | GitHub Copilot | .agents/skills/ | | Kimi Code | .agents/skills/ | | Pi | .pi/skills/ |

The Gemini CLI row is legacy support: Antigravity replaced it, but the installer still writes there for existing setups.

Antigravity, Copilot, Kimi Code, and Amp share one folder. Copilot also reads .github/skills/ and .claude/skills/, and Kimi Code also reads .kimi-code/skills/, but the installer writes the folder they have in common, so picking them together installs antislop once.

Cline, Amp, and Pi each keep a folder of their own and still read one the installer writes. Cline loads .cline/skills/, Amp the shared .agents/skills/, and Pi .pi/skills/, and all three also load a second folder the installer writes: Cline and Amp take .claude/skills/, Pi takes the shared .agents/skills/. So installing Cline or Amp alongside Claude Code, or Pi alongside Antigravity, Copilot, Kimi Code, or Amp, puts the same names in two folders that agent reads, and the installer names that collision. Cline and Amp also resolve a collision that comes from scope in the same unusual direction, and their own documentation is the source: a global skill outranks a project one, so a stale global install silently wins over a fresh project one. Pi settles its collisions the other way round, in code rather than in documentation: project .pi/skills/ beats the shared .agents/skills/, and it beats the user-level ~/.pi/agent/skills/ as well, and a duplicate name is reported with the copy it kept and the copy it dropped. The rest of the agents here do not settle it.

Those are the project paths. A global install writes the same folder under your home directory, with five exceptions: OpenCode writes to ~/.config/opencode/skills/, Antigravity to ~/.gemini/config/skills/, Codex to ~/.agents/skills/, the user-level folder Codex documents in place of its own ~/.codex/skills/, Amp to `~/.con